27980d8e5b5017b296f13761bfbd519a65ec4e68
[libav.git] / libavcodec / utvideodec.c
1 /*
2 * Ut Video decoder
3 * Copyright (c) 2011 Konstantin Shishkov
4 *
5 * This file is part of Libav.
6 *
7 * Libav is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * Libav is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with Libav; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22 /**
23 * @file
24 * Ut Video decoder
25 */
26
27 #include <stdlib.h>
28
29 #include "libavutil/intreadwrite.h"
30 #include "avcodec.h"
31 #include "bytestream.h"
32 #include "get_bits.h"
33 #include "dsputil.h"
34 #include "thread.h"
35
36 enum {
37 PRED_NONE = 0,
38 PRED_LEFT,
39 PRED_GRADIENT,
40 PRED_MEDIAN,
41 };
42
43 typedef struct UtvideoContext {
44 AVCodecContext *avctx;
45 AVFrame pic;
46 DSPContext dsp;
47
48 uint32_t frame_info_size, flags, frame_info;
49 int planes;
50 int slices;
51 int compression;
52 int interlaced;
53 int frame_pred;
54
55 uint8_t *slice_bits;
56 int slice_bits_size;
57 } UtvideoContext;
58
59 typedef struct HuffEntry {
60 uint8_t sym;
61 uint8_t len;
62 } HuffEntry;
63
64 static int huff_cmp(const void *a, const void *b)
65 {
66 const HuffEntry *aa = a, *bb = b;
67 return (aa->len - bb->len)*256 + aa->sym - bb->sym;
68 }
69
70 static int build_huff(const uint8_t *src, VLC *vlc, int *fsym)
71 {
72 int i;
73 HuffEntry he[256];
74 int last;
75 uint32_t codes[256];
76 uint8_t bits[256];
77 uint8_t syms[256];
78 uint32_t code;
79
80 *fsym = -1;
81 for (i = 0; i < 256; i++) {
82 he[i].sym = i;
83 he[i].len = *src++;
84 }
85 qsort(he, 256, sizeof(*he), huff_cmp);
86
87 if (!he[0].len) {
88 *fsym = he[0].sym;
89 return 0;
90 }
91 if (he[0].len > 32)
92 return -1;
93
94 last = 255;
95 while (he[last].len == 255 && last)
96 last--;
97
98 code = 1;
99 for (i = last; i >= 0; i--) {
100 codes[i] = code >> (32 - he[i].len);
101 bits[i] = he[i].len;
102 syms[i] = he[i].sym;
103 code += 0x80000000u >> (he[i].len - 1);
104 }
105
106 return ff_init_vlc_sparse(vlc, FFMIN(he[last].len, 9), last + 1,
107 bits, sizeof(*bits), sizeof(*bits),
108 codes, sizeof(*codes), sizeof(*codes),
109 syms, sizeof(*syms), sizeof(*syms), 0);
110 }
111
112 static int decode_plane(UtvideoContext *c, int plane_no,
113 uint8_t *dst, int step, int stride,
114 int width, int height,
115 const uint8_t *src, int use_pred)
116 {
117 int i, j, slice, pix;
118 int sstart, send;
119 VLC vlc;
120 GetBitContext gb;
121 int prev, fsym;
122 const int cmask = ~(!plane_no && c->avctx->pix_fmt == PIX_FMT_YUV420P);
123
124 if (build_huff(src, &vlc, &fsym)) {
125 av_log(c->avctx, AV_LOG_ERROR, "Cannot build Huffman codes\n");
126 return AVERROR_INVALIDDATA;
127 }
128 if (fsym >= 0) { // build_huff reported a symbol to fill slices with
129 send = 0;
130 for (slice = 0; slice < c->slices; slice++) {
131 uint8_t *dest;
132
133 sstart = send;
134 send = (height * (slice + 1) / c->slices) & cmask;
135 dest = dst + sstart * stride;
136
137 prev = 0x80;
138 for (j = sstart; j < send; j++) {
139 for (i = 0; i < width * step; i += step) {
140 pix = fsym;
141 if (use_pred) {
142 prev += pix;
143 pix = prev;
144 }
145 dest[i] = pix;
146 }
147 dest += stride;
148 }
149 }
150 return 0;
151 }
152
153 src += 256;
154
155 send = 0;
156 for (slice = 0; slice < c->slices; slice++) {
157 uint8_t *dest;
158 int slice_data_start, slice_data_end, slice_size;
159
160 sstart = send;
161 send = (height * (slice + 1) / c->slices) & cmask;
162 dest = dst + sstart * stride;
163
164 // slice offset and size validation was done earlier
165 slice_data_start = slice ? AV_RL32(src + slice * 4 - 4) : 0;
166 slice_data_end = AV_RL32(src + slice * 4);
167 slice_size = slice_data_end - slice_data_start;
168
169 if (!slice_size) {
170 for (j = sstart; j < send; j++) {
171 for (i = 0; i < width * step; i += step)
172 dest[i] = 0x80;
173 dest += stride;
174 }
175 continue;
176 }
177
178 memcpy(c->slice_bits, src + slice_data_start + c->slices * 4,
179 slice_size);
180 memset(c->slice_bits + slice_size, 0, FF_INPUT_BUFFER_PADDING_SIZE);
181 c->dsp.bswap_buf((uint32_t *) c->slice_bits, (uint32_t *) c->slice_bits,
182 (slice_data_end - slice_data_start + 3) >> 2);
183 init_get_bits(&gb, c->slice_bits, slice_size * 8);
184
185 prev = 0x80;
186 for (j = sstart; j < send; j++) {
187 for (i = 0; i < width * step; i += step) {
188 if (get_bits_left(&gb) <= 0) {
189 av_log(c->avctx, AV_LOG_ERROR,
190 "Slice decoding ran out of bits\n");
191 goto fail;
192 }
193 pix = get_vlc2(&gb, vlc.table, vlc.bits, 4);
194 if (pix < 0) {
195 av_log(c->avctx, AV_LOG_ERROR, "Decoding error\n");
196 goto fail;
197 }
198 if (use_pred) {
199 prev += pix;
200 pix = prev;
201 }
202 dest[i] = pix;
203 }
204 dest += stride;
205 }
206 if (get_bits_left(&gb) > 32)
207 av_log(c->avctx, AV_LOG_WARNING,
208 "%d bits left after decoding slice\n", get_bits_left(&gb));
209 }
210
211 ff_free_vlc(&vlc);
212
213 return 0;
214 fail:
215 ff_free_vlc(&vlc);
216 return AVERROR_INVALIDDATA;
217 }
218
219 static const int rgb_order[4] = { 1, 2, 0, 3 };
220
221 static void restore_rgb_planes(uint8_t *src, int step, int stride, int width,
222 int height)
223 {
224 int i, j;
225 uint8_t r, g, b;
226
227 for (j = 0; j < height; j++) {
228 for (i = 0; i < width * step; i += step) {
229 r = src[i];
230 g = src[i + 1];
231 b = src[i + 2];
232 src[i] = r + g - 0x80;
233 src[i + 2] = b + g - 0x80;
234 }
235 src += stride;
236 }
237 }
238
239 static void restore_median(uint8_t *src, int step, int stride,
240 int width, int height, int slices, int rmode)
241 {
242 int i, j, slice;
243 int A, B, C;
244 uint8_t *bsrc;
245 int slice_start, slice_height;
246 const int cmask = ~rmode;
247
248 for (slice = 0; slice < slices; slice++) {
249 slice_start = ((slice * height) / slices) & cmask;
250 slice_height = ((((slice + 1) * height) / slices) & cmask) -
251 slice_start;
252
253 bsrc = src + slice_start * stride;
254
255 // first line - left neighbour prediction
256 bsrc[0] += 0x80;
257 A = bsrc[0];
258 for (i = step; i < width * step; i += step) {
259 bsrc[i] += A;
260 A = bsrc[i];
261 }
262 bsrc += stride;
263 if (slice_height == 1)
264 continue;
265 // second line - first element has top prediction, the rest uses median
266 C = bsrc[-stride];
267 bsrc[0] += C;
268 A = bsrc[0];
269 for (i = step; i < width * step; i += step) {
270 B = bsrc[i - stride];
271 bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
272 C = B;
273 A = bsrc[i];
274 }
275 bsrc += stride;
276 // the rest of lines use continuous median prediction
277 for (j = 2; j < slice_height; j++) {
278 for (i = 0; i < width * step; i += step) {
279 B = bsrc[i - stride];
280 bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
281 C = B;
282 A = bsrc[i];
283 }
284 bsrc += stride;
285 }
286 }
287 }
288
289 /* UtVideo interlaced mode treats every two lines as a single one,
290 * so restoring function should take care of possible padding between
291 * two parts of the same "line".
292 */
293 static void restore_median_il(uint8_t *src, int step, int stride,
294 int width, int height, int slices, int rmode)
295 {
296 int i, j, slice;
297 int A, B, C;
298 uint8_t *bsrc;
299 int slice_start, slice_height;
300 const int cmask = ~(rmode ? 3 : 1);
301 const int stride2 = stride << 1;
302
303 for (slice = 0; slice < slices; slice++) {
304 slice_start = ((slice * height) / slices) & cmask;
305 slice_height = ((((slice + 1) * height) / slices) & cmask) -
306 slice_start;
307 slice_height >>= 1;
308
309 bsrc = src + slice_start * stride;
310
311 // first line - left neighbour prediction
312 bsrc[0] += 0x80;
313 A = bsrc[0];
314 for (i = step; i < width * step; i += step) {
315 bsrc[i] += A;
316 A = bsrc[i];
317 }
318 for (i = 0; i < width * step; i += step) {
319 bsrc[stride + i] += A;
320 A = bsrc[stride + i];
321 }
322 bsrc += stride2;
323 if (slice_height == 1)
324 continue;
325 // second line - first element has top prediction, the rest uses median
326 C = bsrc[-stride2];
327 bsrc[0] += C;
328 A = bsrc[0];
329 for (i = step; i < width * step; i += step) {
330 B = bsrc[i - stride2];
331 bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
332 C = B;
333 A = bsrc[i];
334 }
335 for (i = 0; i < width * step; i += step) {
336 B = bsrc[i - stride];
337 bsrc[stride + i] += mid_pred(A, B, (uint8_t)(A + B - C));
338 C = B;
339 A = bsrc[stride + i];
340 }
341 bsrc += stride2;
342 // the rest of lines use continuous median prediction
343 for (j = 2; j < slice_height; j++) {
344 for (i = 0; i < width * step; i += step) {
345 B = bsrc[i - stride2];
346 bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
347 C = B;
348 A = bsrc[i];
349 }
350 for (i = 0; i < width * step; i += step) {
351 B = bsrc[i - stride];
352 bsrc[i + stride] += mid_pred(A, B, (uint8_t)(A + B - C));
353 C = B;
354 A = bsrc[i + stride];
355 }
356 bsrc += stride2;
357 }
358 }
359 }
360
361 static int decode_frame(AVCodecContext *avctx, void *data, int *data_size,
362 AVPacket *avpkt)
363 {
364 const uint8_t *buf = avpkt->data;
365 int buf_size = avpkt->size;
366 UtvideoContext *c = avctx->priv_data;
367 int i, j;
368 const uint8_t *plane_start[5];
369 int plane_size, max_slice_size = 0, slice_start, slice_end, slice_size;
370 int ret;
371 GetByteContext gb;
372
373 if (c->pic.data[0])
374 ff_thread_release_buffer(avctx, &c->pic);
375
376 c->pic.reference = 1;
377 c->pic.buffer_hints = FF_BUFFER_HINTS_VALID;
378 if ((ret = ff_thread_get_buffer(avctx, &c->pic)) < 0) {
379 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
380 return ret;
381 }
382
383 ff_thread_finish_setup(avctx);
384
385 /* parse plane structure to get frame flags and validate slice offsets */
386 bytestream2_init(&gb, buf, buf_size);
387 for (i = 0; i < c->planes; i++) {
388 plane_start[i] = gb.buffer;
389 if (bytestream2_get_bytes_left(&gb) < 256 + 4 * c->slices) {
390 av_log(avctx, AV_LOG_ERROR, "Insufficient data for a plane\n");
391 return AVERROR_INVALIDDATA;
392 }
393 bytestream2_skipu(&gb, 256);
394 slice_start = 0;
395 slice_end = 0;
396 for (j = 0; j < c->slices; j++) {
397 slice_end = bytestream2_get_le32u(&gb);
398 slice_size = slice_end - slice_start;
399 if (slice_end <= 0 || slice_size <= 0 ||
400 bytestream2_get_bytes_left(&gb) < slice_end) {
401 av_log(avctx, AV_LOG_ERROR, "Incorrect slice size\n");
402 return AVERROR_INVALIDDATA;
403 }
404 slice_start = slice_end;
405 max_slice_size = FFMAX(max_slice_size, slice_size);
406 }
407 plane_size = slice_end;
408 bytestream2_skipu(&gb, plane_size);
409 }
410 plane_start[c->planes] = gb.buffer;
411 if (bytestream2_get_bytes_left(&gb) < c->frame_info_size) {
412 av_log(avctx, AV_LOG_ERROR, "Not enough data for frame information\n");
413 return AVERROR_INVALIDDATA;
414 }
415 c->frame_info = bytestream2_get_le32u(&gb);
416 av_log(avctx, AV_LOG_DEBUG, "frame information flags %X\n", c->frame_info);
417
418 c->frame_pred = (c->frame_info >> 8) & 3;
419
420 if (c->frame_pred == PRED_GRADIENT) {
421 av_log_ask_for_sample(avctx, "Frame uses gradient prediction\n");
422 return AVERROR_PATCHWELCOME;
423 }
424
425 av_fast_malloc(&c->slice_bits, &c->slice_bits_size,
426 max_slice_size + FF_INPUT_BUFFER_PADDING_SIZE);
427
428 if (!c->slice_bits) {
429 av_log(avctx, AV_LOG_ERROR, "Cannot allocate temporary buffer\n");
430 return AVERROR(ENOMEM);
431 }
432
433 switch (c->avctx->pix_fmt) {
434 case PIX_FMT_RGB24:
435 case PIX_FMT_RGBA:
436 for (i = 0; i < c->planes; i++) {
437 ret = decode_plane(c, i, c->pic.data[0] + rgb_order[i], c->planes,
438 c->pic.linesize[0], avctx->width, avctx->height,
439 plane_start[i], c->frame_pred == PRED_LEFT);
440 if (ret)
441 return ret;
442 if (c->frame_pred == PRED_MEDIAN) {
443 if (!c->interlaced) {
444 restore_median(c->pic.data[0] + rgb_order[i], c->planes,
445 c->pic.linesize[0], avctx->width,
446 avctx->height, c->slices, 0);
447 } else {
448 restore_median_il(c->pic.data[0] + rgb_order[i], c->planes,
449 c->pic.linesize[0], avctx->width,
450 avctx->height, c->slices, 0);
451 }
452 }
453 }
454 restore_rgb_planes(c->pic.data[0], c->planes, c->pic.linesize[0],
455 avctx->width, avctx->height);
456 break;
457 case PIX_FMT_YUV420P:
458 for (i = 0; i < 3; i++) {
459 ret = decode_plane(c, i, c->pic.data[i], 1, c->pic.linesize[i],
460 avctx->width >> !!i, avctx->height >> !!i,
461 plane_start[i], c->frame_pred == PRED_LEFT);
462 if (ret)
463 return ret;
464 if (c->frame_pred == PRED_MEDIAN) {
465 if (!c->interlaced) {
466 restore_median(c->pic.data[i], 1, c->pic.linesize[i],
467 avctx->width >> !!i, avctx->height >> !!i,
468 c->slices, !i);
469 } else {
470 restore_median_il(c->pic.data[i], 1, c->pic.linesize[i],
471 avctx->width >> !!i,
472 avctx->height >> !!i,
473 c->slices, !i);
474 }
475 }
476 }
477 break;
478 case PIX_FMT_YUV422P:
479 for (i = 0; i < 3; i++) {
480 ret = decode_plane(c, i, c->pic.data[i], 1, c->pic.linesize[i],
481 avctx->width >> !!i, avctx->height,
482 plane_start[i], c->frame_pred == PRED_LEFT);
483 if (ret)
484 return ret;
485 if (c->frame_pred == PRED_MEDIAN) {
486 if (!c->interlaced) {
487 restore_median(c->pic.data[i], 1, c->pic.linesize[i],
488 avctx->width >> !!i, avctx->height,
489 c->slices, 0);
490 } else {
491 restore_median_il(c->pic.data[i], 1, c->pic.linesize[i],
492 avctx->width >> !!i, avctx->height,
493 c->slices, 0);
494 }
495 }
496 }
497 break;
498 }
499
500 c->pic.key_frame = 1;
501 c->pic.pict_type = AV_PICTURE_TYPE_I;
502 c->pic.interlaced_frame = !!c->interlaced;
503
504 *data_size = sizeof(AVFrame);
505 *(AVFrame*)data = c->pic;
506
507 /* always report that the buffer was completely consumed */
508 return buf_size;
509 }
510
511 static av_cold int decode_init(AVCodecContext *avctx)
512 {
513 UtvideoContext * const c = avctx->priv_data;
514
515 c->avctx = avctx;
516
517 ff_dsputil_init(&c->dsp, avctx);
518
519 if (avctx->extradata_size < 16) {
520 av_log(avctx, AV_LOG_ERROR,
521 "Insufficient extradata size %d, should be at least 16\n",
522 avctx->extradata_size);
523 return AVERROR_INVALIDDATA;
524 }
525
526 av_log(avctx, AV_LOG_DEBUG, "Encoder version %d.%d.%d.%d\n",
527 avctx->extradata[3], avctx->extradata[2],
528 avctx->extradata[1], avctx->extradata[0]);
529 av_log(avctx, AV_LOG_DEBUG, "Original format %X\n",
530 AV_RB32(avctx->extradata + 4));
531 c->frame_info_size = AV_RL32(avctx->extradata + 8);
532 c->flags = AV_RL32(avctx->extradata + 12);
533
534 if (c->frame_info_size != 4)
535 av_log_ask_for_sample(avctx, "Frame info is not 4 bytes\n");
536 av_log(avctx, AV_LOG_DEBUG, "Encoding parameters %08X\n", c->flags);
537 c->slices = (c->flags >> 24) + 1;
538 c->compression = c->flags & 1;
539 c->interlaced = c->flags & 0x800;
540
541 c->slice_bits_size = 0;
542
543 switch (avctx->codec_tag) {
544 case MKTAG('U', 'L', 'R', 'G'):
545 c->planes = 3;
546 avctx->pix_fmt = PIX_FMT_RGB24;
547 break;
548 case MKTAG('U', 'L', 'R', 'A'):
549 c->planes = 4;
550 avctx->pix_fmt = PIX_FMT_RGBA;
551 break;
552 case MKTAG('U', 'L', 'Y', '0'):
553 c->planes = 3;
554 avctx->pix_fmt = PIX_FMT_YUV420P;
555 break;
556 case MKTAG('U', 'L', 'Y', '2'):
557 c->planes = 3;
558 avctx->pix_fmt = PIX_FMT_YUV422P;
559 break;
560 default:
561 av_log(avctx, AV_LOG_ERROR, "Unknown Ut Video FOURCC provided (%08X)\n",
562 avctx->codec_tag);
563 return AVERROR_INVALIDDATA;
564 }
565
566 return 0;
567 }
568
569 static av_cold int decode_end(AVCodecContext *avctx)
570 {
571 UtvideoContext * const c = avctx->priv_data;
572
573 if (c->pic.data[0])
574 ff_thread_release_buffer(avctx, &c->pic);
575
576 av_freep(&c->slice_bits);
577
578 return 0;
579 }
580
581 AVCodec ff_utvideo_decoder = {
582 .name = "utvideo",
583 .type = AVMEDIA_TYPE_VIDEO,
584 .id = AV_CODEC_ID_UTVIDEO,
585 .priv_data_size = sizeof(UtvideoContext),
586 .init = decode_init,
587 .close = decode_end,
588 .decode = decode_frame,
589 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_FRAME_THREADS,
590 .long_name = NULL_IF_CONFIG_SMALL("Ut Video"),
591 };